Amazon's cloud now less prone to failure
EC2 adds more elasticity and zonality
Amazon's utility computing-style EC2 service just got more muscular. Thanks to a pair of new options, customers can take steps to ensure their applications keep running when system failures or network disruptions occur.
First off, Amazon has rolled out "Availability Zones", which address a major weakness in the EC2 system. Customers have been able to run multiple copies of their applications within Amazon's EC2 computing centers. The problem, however, was that Amazon planted those instances wherever its system saw fit. So, a customer could theoretically end up running multiple copies of its applications on a single physical server, which rather defeats the redundancy goal.
Through an API call, customers can now direct their applications to specific systems or availability zones. Each of the zones has its own power, networking and cooling systems, which means that your application should stay up if a major issue such as a fire or network outage affects one zone.
There's more detailed information on how to configure your zones here.
In addition, Amazon will offer customers "Elastic IP" addresses, which are cloudified versions of static IP addresses. Basically, you can now tie a given IP address to your overall account rather than a specific server. Amazon thinks this gives customers more options when dealing with potential system failures or just tweaks to a given a configuration.
For more, we turn to the horse's mouth,
Unlike traditional static IP addresses, Elastic IP addresses can be dynamically remapped on the fly to point to any compute instance in a developer’s Amazon EC2 account. This means that rather than waiting on a data technician to reconfigure or replace a host, or waiting for DNS to propagate to all of their customers, developers can now engineer around problems with their instance or software by quickly remapping their Elastic IP address to a replacement instance. Elastic IP addresses make it easy for companies to host websites, web services, and other online applications on Amazon EC2, enabling a new range of customers to take advantage of Amazon’s elastic, on-demand, cloud computing offering.
There's additional bits on the Elastic IPs here.
Digging deeper into EC2, developers will also find that Amazon has added more flexibility around the creation of AMIs (Amazon Machine Images), which are the actual software packages running on the service.
We're told that customers can "select the kernel and RAM disk to bundle with your AMI or to specify a kernel and RAM disk at launch time."
Some new AMIs include:
- Fedora Core 6 - 32 bit - a stock FC6 release with matching kernel and RAM disk
- Fedora 8 - 32 bit - a stock F8 release with matching kernel and RAM disk
- Fedora 8 - 64 bit - a stock F8 release with matching kernel and RAM disk
- 2.6.18 Kernel - 32 bit - a stock 2.6.18 kernel (can be used with 32 bit AMIs)
- 2.6.18 Kernel - 64 bit - a stock 2.6.18 kernel (can be used with 64 bit AMIs)
The major server vendors, excluding a lurching Sun Microsystems, appear content to let Amazon roll out and now nicely refine its utility computing service. How curious.
With Amazon being the major game in town, customers must be pleased to see it add in these types of resiliency features. It's only worth putting up with the 'cloud computing' lingo if the technology actually lives up to its billing and reduces some of the headaches associated with relying on dedicated, physical systems. ®
Nothing new about Elastic IPs
There's nothing new in the Elastic IP deal apart from how they charge for it.
Compaq Tru64 clusters had "application IPs" which moved from node to node as the resources were moved. We were using this technology 10 or more years ago.
Paris, because there's nothing new in this world...
Service vendors are sweating, not waiting
I make it a point to tell our colo providers, who charge by the rack-unit, how completely satisfied we are with the utility computing model. Our accounting department loves it too.
EC2 now is like Linux was for us a few years ago... All new systems are planned and provisioned for this new model, and mentioning it to our legacy vendors usually gets a discount from the sales droids that recognize EC2 as competition.
I've still got pleasant tingles from the IT headcount reductions that virtualization got us a few years ago, and it makes me drool to think that I can do it again with utility computing.